HOI4 - Development Diary - 19th Of August 2016

[CaosRL]

I would like to be able to create puppets in occupied territory

 

[BjornB]

Thanks to everyone posting constructive feedback!

A reminder for some (only a few) others to watch and learn the proper way to give constructive feedback.

 

[tom_jones]

The tests we have done here have been looking good.

What do you mean by this? The FRA and ENG actually edit their infantry templates, the AI doesn't wildly exceed the desired width, inserts battalions and support units in accordance with supplied target_template, the AI pays attention to supplied attribute weights?

If that's the case, have you don't some changes to the relevant code, or is this with thing the way they were?

(tl;dr do you mean "we changed the code and it now seems fixed" vs "we did a check and it worked for us so there's nothing to fix")

 

[Ricard2037]

BjornB About the submarines, convoy raiding, resources and trade... it's working as expected, or you expect to make some changes on the next patches.

At my experience with germany, or submarines must be upgraded, or the convoys are broken, but on the other side, the trade convoys work fine!!
Thanks =D

 

[SteelVolt]

What do you mean by this? The FRA and ENG actually edit their infantry templates, the AI doesn't wildly exceed the desired width, inserts battalions and support units in accordance with supplied target_template, the AI pays attention to supplied attribute weights?

If that's the case, have you don't some changes to the relevant code, or is this with thing the way they were?

(tl;dr do you mean "we changed the code and it now seems fixed" vs "we did a check and it worked for us so there's nothing to fix")

The template designs we are seeing are fairly good so it is not something we are looking closer at currently, because we have bigger fish to fry. As far as I have been able to tell it follows the target template well enough, and moves on as intended with the progression scripts.

 

[Meglok]

BjornB About the submarines, convoy raiding, resources and trade... it's working as expected, or you expect to make some changes on the next patches.

At my experience with germany, or submarines must be upgraded, or the convoys are broken, but on the other side, the trade convoys work fine!!
Thanks =D

See this thread https://forum.paradoxplaza.com/foru...in-the-battle-of-the-atlantic-op-edit.962198/

PDS is aware of the issues with convoys etc.

 

[seattle]

The template designs we are seeing are fairly good so it is not something we are looking closer at currently, because we have bigger fish to fry. As far as I have been able to tell it follows the target template well enough, and moves on as intended with the progression scripts.

However, aren't most templates way too small? A German infantry division consisted of 17,000 men initially which is roughly width 40 in HoI4 terms (incl. arty etc.).
Most templates I've seen when playing a minor nation were tiny, like 4,000-5,000 men.
This leads to minors having 3 times the number of units of the map that they probably should have. ==> major cause of the current performance issues ==> easy fix: use bigger, more competitive, more realistic templates.

 

[tom_jones]

The template designs we are seeing are fairly good so it is not something we are looking closer at currently, because we have bigger fish to fry. As far as I have been able to tell it follows the target template well enough, and moves on as intended with the progression scripts.

Sorry, but... how can you say it follows the target template "well enough" when, in a test case where you give it a template with multiple AT battalions, this target is basically ignored and it's a rare occurrence if over long years even one AT is added? (with the AT researched) How is it following "well enough" when you have tank division template with target of even amount of tanks and infantry, but the game just sticks more and more tank battalions into it (driving itself in production backlog black hole) and no infantry whatsoever? How does it follow it well enough when some major nations leave their infantry divisions unchanged from the starting template for literally years, even after WW2 breaks out? How does it follow it well enough when it doesn't insert self-propelled battalions in the mechanized units, even if you instruct the AI to research them (and it does), unlock them in the template settings and put them in the target?

"just ignore the target and do whatever you want, which for regular infantry luckily happens to produce a semi-useful template, but botches everything else" is far from well enough, imo.

I really hope this new progression Band-Aid functions and the regular template code isn't given much opportunity to mess the provided stuff up, because at the moment said regular template code appears badly broken. And that hopefully you can have a look into it after you fry these bigger fish, because that's imo pretty large fish that severely affects how well the AI performs in the combat, too

 

[SteelVolt]

However, aren't most templates way too small? A German infantry division consisted of 17,000 men initially which is roughly width 40 in HoI4 terms (incl. arty etc.).
Most templates I've seen when playing a minor nation were tiny, like 4,000-5,000 men.
This leads to minors having 3 times the number of units of the map that they probably should have. ==> major cause of the current performance issues ==> easy fix: use bigger, more competitive, more realistic templates.

This is one instance where the AI is allowed to 'play the game' rather than aiming towards history. While it is true that the AI needs to be better at managing sizes of templates, this is mainly down to resources vs needs rather than what they looked like historically. Having enough divisions to cover large fronts is a factor for several countries, particularly the majors. Currently this is controlled through scipt rather than dynamically, so the sizes for target templates of infantry divisions can be edited per country.

Most infantry divisions I see in 1945 apear to have +8.5k manpower, though.

Either way, diving on to scripting more personalized target templates for various countries is something I hope to get in to when the bigger issues are taken care of.

 

[SteelVolt]

How does it follow it well enough when some major nations leave their infantry divisions unchanged from the starting template for literally years, even after WW2 breaks out?

This is partly random, so it may happen. Though I have not seen this being the case other than very rarely. If you have a case where you see this a lot it would be helpful if you reported it.

in a test case where you give it a template with multiple AT battalions, this target is basically ignored and it's a rare occurrence if over long years even one AT is added?

It is true that it does not like AT. While I have not had the time or opportunity to investigate, my guess is that the stats for it look so bad that it may take some drastic scripting, such as this:
target_template = {
weight = 1.0

Which would make it completely ignore stats.

but the game just sticks more and more tank battalions into it (driving itself in production backlog black hole) and no infantry whatsoever?

Have not seen the AI go pure armored in a long time, though it tends to go with motorized as it avoids the rather nasty effect on speed that regular infantry gives it. Again; try setting the weight lower. I think the way it is coded makes the stats still matter too much.

How does it follow it well enough when it doesn't insert self-propelled battalions in the mechanized units, even if you instruct the AI to research them (and it does), unlock them in the template settings and put them in the target?

Again, probably a case of stats having a too big effect.

"just ignore the target and do whatever you want, which for regular infantry luckily happens to produce a semi-useful template, but botches everything else" is far from well enough, imo.

I agree that it would be, but disagree that it is as bad as that.

And that hopefully you can have a look into it after you fry these bigger fish, because that's imo pretty large fish that severely affects how well the AI performs in the combat, too

I hope so too.

tl;dr: Well enough because the big typical cases for majors are not totally fubar at the moment (which they realluy reqally have been), particularly early to mid game. Perhaps I should have used the term "well enough, for now". It still needs work to be properly good, especially for late game, and the template progression is the first step towards that.

 

[SteelVolt]

Actually, in the light of the previous comment, I took the liberty to add a field in the AI template scripting that should make it easier to make the target template matter more (or less), plus update the code with it.
Will need to testing, of course.

EDIT: Most notably it should now accept even what it would consider to be negative stats

 

[Meglok]

Actually, in the light of the previous comment, I took the liberty to add a field in the AI template scripting that should make it easier to make the target template matter more (or less), plus update the code with it.
Will need to testing, of course.

EDIT: Most notably it should now accept even what it would consider to be negative stats

Thank you!

 

[tom_jones]

This is partly random, so it may happen. Though I have not seen this being the case other than very rarely. If you have a case where you see this a lot it would be helpful if you reported it.

I see this happen literally 100% of the time with ENG and FRA, in the hands-off AI tests I run. But it makes for a poor bug report because "steps to reproduce: start the game and let it play, have a look at FRA and ENG" doesn't mean much if you don't evidence this on your end.

It is true that it does not like AT. While I have not had the time or opportunity to investigate, my guess is that the stats for it look so bad that it may take some drastic scripting, such as this:
target_template = {
weight = 1.0

Which would make it completely ignore stats.

I have tried different ways to coax the AI to follow the target, I've documented some of them in the report here Even if you reduce the weights to the point where they should be no factor, or drive the target weight up, or both, the AI behavior doesn't change. If you set the weights in a manner that *should* have impact in very specific way, they don't. Basically, I don't know what the template code is doing, but it sure as heck doesn't appear to pay much attention to either target template *or* weights. As well as other parameters, like target width etc.

Have not seen the AI go pure armored in a long time, though it tends to go with motorized as it avoids the rather nasty effect on speed that regular infantry gives it. Again; try setting the weight lower. I think the way it is coded makes the stats still matter too much.

What I mean is the AI keeps adding tank battalions to the template which has some amount of infantry and tanks at the start, but it doesn't add any extra infantry, including the motorized one. This is, again, with various combinations of stats and/or weights. Plus honestly, given the target weight of light_armour is set to 0.9 by default and it still exhibits this, why would the stats still matter so much in such case? Wouldn't you agree that's a faulty behavior, too?

(if I had to guess, this may be driven by the fact the target template has tank battalion as first entry and infantry battalions are listed next. And when the code pads the template with extra units it just grabs the first entry from the target and runs with it? But i'd have to test if flipping the order around changes that)

 

[rip4ever]

It is b.c. its micro Hell - and unituitive - the decisions of actual Inflight WIngsizes should not be made on the Top Level (aka the Player) but on the Tactical Level (aka the Simulation)

So should the decision of whether the plane flies a mission over New York or the entire east coast!!!!

 

[SteelVolt]

I see this happen literally 100% of the time with ENG and FRA, in the hands-off AI tests I run. But it makes for a poor bug report because "steps to reproduce: start the game and let it play, have a look at FRA and ENG" doesn't mean much if you don't evidence this on your end.

I will try to remember to have a look at it tomorrow after an over night (which is a good wat to reproduce such issues.)

I have tried different ways to coax the AI to follow the target, I've documented them in the report here Even if you reduce the weights to the point where they should be no factor, or drive the target weight up, or both, the AI behavior doesn't change. If you set the weights in a manner that *should* have impact in very specific way, they don't. Basically, I don't know what the template code is doing, but it sure as heck doesn't appear to pay much attention to either target template *or* weights. As well as other parameters, like target width etc.

Then it is in the pipeline and will arrive at my table eventually. Sure does sound like a bug.

What I mean is the AI keeps adding tank battalions to the template which has some amount of infantry and tanks at the start, but it doesn't add any extra infantry, including the motorized one. This is, again, with various combinations of stats and/or weights. Plus honestly, given the target weight of light_armour is set to 0.9 by default and it still exhibits this, why would the stats still matter so much in such case? Wouldn't you agree that's a faulty behavior, too?

I would agree that it is faulty behaviour, but still 'well enough' in the sense that the AI manages to produce usable templates in the most important cases.

And when the code pads the template with extra units it just grabs the first entry from the target and runs with it?

Unlikely, given that I consistently see the AI replacing a couple of infantry battalions with artillery.

 

[tom_jones]

Unlikely, given that I consistently see the AI replacing a couple of infantry battalions with artillery.

That's a good point; must be something else, wish I knew what, though.

Incidentally, I've ran another game in the meantime, with the target template weight set to 1.0 so in theory attribute weights shouldn't matter and it should follow the target exactly, right? Well it's April 1937 and Italy already shows it doesn't give any fucks:

(target is 8 inf, 1 art, 1 at. target width is 19.9)

so... yeah, I don't know but it sure doesn't look like it's working the way it's supposed to work.

if it's some randomness factor at play, please either disable it or expose it in the template config, so it can be decided there just how random the result should be..?

 

[seattle]

This is one instance where the AI is allowed to 'play the game' rather than aiming towards history. While it is true that the AI needs to be better at managing sizes of templates, this is mainly down to resources vs needs rather than what they looked like historically. Having enough divisions to cover large fronts is a factor for several countries, particularly the majors. Currently this is controlled through scipt rather than dynamically, so the sizes for target templates of infantry divisions can be edited per country.

Most infantry divisions I see in 1945 apear to have +8.5k manpower, though.

Either way, diving on to scripting more personalized target templates for various countries is something I hope to get in to when the bigger issues are taken care of.

That's where I wholeheartedly disagree!
As Germany I am quite capable of blitzing through WW2 with a total of ~150 units. Barbarossa needs no more than 100 units without holes in the frontline.
I fail to see how Romania, Nat Spain, Hungary etc. need 150+ units to cover their extremely limited frontlines.
I also fail to see how Germany needs 500+ units...

I'm pretty sure that all performance issues would be history if the total number of units on the map was halved. It's not about historical railroading, but rather CPU performance and a.i. issues.

 

[freespace2]

Dev I hope you all enjoyed your vacation but when are you planning to fix more important issues like Multiplayer and AI. I hope this is not the full patch note for 1.3 as it seems to be a very minor improvement which ignore more critical issues.

 

[Broletariat90]

Dev I hope you all enjoyed your vacation but when are you planning to fix more important issues like Multiplayer and AI. I hope this is not the full patch note for 1.3 as it seems to be a very minor improvement which ignore more critical issues.

They addressed AI in last Friday's diary. And they have stated it multiple times in threads. They know what needs to be done. They also have another 3 weeks of work to do.

 

[Jamey]

Sorry about not being able to respond to all your questions. Some of them were better for Podcat to look at and I thought he would be back by now.
Anyway, I am off to have me some weekend, and I hope you all enjoy yours

I just got done catching up with all of the dev posts here. I hope you had a good weekend, and I look forward to giving a look at the results of all of these improvements when they are available in a beta.